Hi everyone,
I have a question regarding the HA quorum behavior of a Proxmox VE cluster in a multi-site environment.
My current architecture is as shown in the diagrams:
The cluster has a total of 11 votes. Additionally, all VMs across nodes utilize shared storage, with data redundancy synchronization between sites . This ensures that even after a site network disconnection, the storage in each location still maintains complete VM images.
The challenge is: when the networks between the three sites are completely severed (indicated by the red lines), but the internal network within each DC remains functional, DC-A only has 5 votes and DC-B also has 5 votes. Since neither reaches the 6-vote majority, all nodes in both DC-A and DC-B trigger fencing and reboot via the watchdog.
My goal is to designate DC-A as the "privileged majority partition" without adding physical nodes. I want to ensure that DC-A nodes remain operational and avoid fencing during a network partition, while automatically restarting DC-B's VMs via HA.
What is the recommended way to implement this correctly in PVE?
I have a question regarding the HA quorum behavior of a Proxmox VE cluster in a multi-site environment.
My current architecture is as shown in the diagrams:
- DC-A: 5 nodes (node-a1 to node-a5)
- DC-B: 5 nodes (node-b1 to node-b5)
- DC-C: 1 QDevice (External Voter)
The cluster has a total of 11 votes. Additionally, all VMs across nodes utilize shared storage, with data redundancy synchronization between sites . This ensures that even after a site network disconnection, the storage in each location still maintains complete VM images.
The challenge is: when the networks between the three sites are completely severed (indicated by the red lines), but the internal network within each DC remains functional, DC-A only has 5 votes and DC-B also has 5 votes. Since neither reaches the 6-vote majority, all nodes in both DC-A and DC-B trigger fencing and reboot via the watchdog.
My goal is to designate DC-A as the "privileged majority partition" without adding physical nodes. I want to ensure that DC-A nodes remain operational and avoid fencing during a network partition, while automatically restarting DC-B's VMs via HA.
What is the recommended way to implement this correctly in PVE?
Last edited: